for the parsing filter primarily involved algorithmic detection. This algorithmic parsing method is essentially nonlinear, although we recognized as time progressed that it would be more efficient to employ a method implementing and exploiting the advantages of both the linear and the nonlinear methods.
With regards to the particulars of implementation, the algorithm would first establish a rolling average and rolling average slope. If the current ECG value were much greater than the average and the current slope much greater or less than the recent average slope, the value would be identified as a possible heartbeat, and an initial heartbeat could be established if the difference between the ECG and the average was above a certain threshold.
With a pair of established heartbeats, the algorithm would determine the heart rate and would predict when the next heartbeat would occur. Knowing the absolute limits of the heart rate of the animal that produced the ECG, the algorithm would then find all the possible heartbeats within this range, and the possible heartbeat within this range closest to the predicted heartbeat time and largest in ECG value would be identified as a heartbeat. The new averaged heart rate would be determined, and this process of determining the next heartbeat would start over again.
A later version of the program instead took the view that sometimes noise would inhibit the algorithm (and human viewers) from choosing the correct heartbeat, and therefore this segment should simply be discarded. A real-time boxcar-like filter was set up to determine when noise was ineffectual enough to continue with the algorithm, and when to stop until definite heartbeats could be established again. This feature would essentially filter out "bad" ECG data while retaining and analyzing the rest of the ECG data.
